[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-565912-105":53,"doc-detail-565912-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","stool-classication-using-deep-metric-learning","Stool Classification Using Deep Metric Learning","","Chronic gastrointestinal disorders are common in the U.S. but are hard to measure with convenient and accurate tools, motivating an image-based system to extract clinical signals from stool photographs. The work crowdsourced a stool dataset in the wild and annotated it with the Bristol Stool Scale (BSS), then trained a ResNet18 classifier and explored deep metric embeddings using triplet and contrastive loss for low-shot categories. Results show promising mean per-class absolute deviation of 0.82 BSS and improved robustness, though metric learning did not surpass the ResNet18, likely due to limited unlabeled data.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":11,"@type":70,"position":76},"https://docshare.wps.com/template/presentations/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/stool-classication-using-deep-metric-learning/565912/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/stool-classication-using-deep-metric-learning/565912.png","ImageObject",442,249,{"name":88,"@type":89},"WPS_1786070896","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-10-19","2026-10-13",true,{"@type":98,"interactionType":99,"userInteractionCount":9},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"What problem does the paper address?","Question",{"text":108,"@type":109},"The paper targets the lack of convenient, accurate measurement tools for chronic gut disorders by using stool images to extract standardized clinical data.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"How is the Bristol Stool Scale (BSS) used in the system?",{"text":113,"@type":109},"Physicians assign a BSS score on a 1–7 scale based on visual assessment, and the model predicts BSS to quantify stool consistency over time.",{"name":115,"@type":106,"acceptedAnswer":116},"What models and training strategies are evaluated?",{"text":117,"@type":109},"A ResNet18 baseline is trained to predict BSS, and deep metric learning is tested using triplet and contrastive loss by turning the CNN into a feature extractor and optimizing embeddings.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},565912,1792382267,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":8,"category_name":11,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":9,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":135,"language":136,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":137,"faqs":138,"seo_title":139,"seo_description":61,"update_tm":140,"read_time":76},549768072016,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","Stool Classification Using Deep Metric Learning  \nEric Nguyen Cornell Tech  \nM.Eng. Computer Science [en274@cornell.edu](en274@cornell.edu)  \nChris Kruger Cornell Tech  \nM.Eng. Computer Science [crk78@cornell.edu](crk78@cornell.edu)  \nSkyler Erickson Cornell Tech  \nM.Eng. Computer Science [mse53@cornell.edu](mse53@cornell.edu)  \nAbstract  \nChronic gut disorders in the U.S. are commonplace, however, they lack convenient and accurate measurement tools. We propose a system that allows a person to take a picture of their stool and extract clinical data from the image to track their symptoms. To train this system, we crowdsourced a stool dataset in the wild through [http://train.auggi.ai and had physicians annotate](http://train.auggi.ai and had physicians annotate) them with the Bristol Stool Scale (BSS) . The dataset gathered is highly imbalanced, with type 1 having just 6 images. We trained a ResNet18 classifier to predict the BSS, and then used deep metric embeddings (triplet and contrastive loss) with the goal of improving the performance of the low-shot classes. Triplet and contrastive loss are evaluated as potential improvements on our baseline ResNet18 models due to their additional information beyond purely vision. Triplet loss uses an anchor, positive, and negative to push similar images closer and dissimilar images further apart. Contrastive applies the same concept with pairs instead of triplets. Our ResNet18 trained model showed promising results with a mean per class absolute deviation of 0 .82 BSS, which means on average, the model was within one BSS value from the true label, reasonable for clinical applications and a large improvement over what patients are able to do on their own. Compared to GI physicians, our deep learning models were not as accurate. Amongst the deep learning techniques, the ResNet18 performed the best, while the metric learning models were not able to improve performance over theResNet18 . This may be due to the lack of unlabeled data that help the metric models create embeddings that best separate the classes. With a modest increase in the data in these few labeled classes, we expect the performance to significantly improve and be more in-line with what doctors are able to predict from stool images.  \n1. Introduction  \nOne in five Americans suffer from some form of chronic gastrointestinal (GI) disorder [1] . Managing treatment for  \nthese disorders, such as irritable bowel syndrome, often requires tracking symptoms and lifestyle patterns over time. These patient-reported inputs are often subjective and make it difficult to control their symptoms. We propose a model that can allow patients to take a picture of their stool and automatically extract clinical data from the image. This allows patients to track the physical manifestation of their disorder over time in a convenient way. This system also allows doctors to view objective and standardized metrics about the state of their patient's digestive system.  \nOne standardized metric doctors use for GI health track  \ning is the Bristol Stool Scale (BSS) . Based on a visual assessment, doctors assign a score on a 1 to 7 point scale to quantify the consistency of a patient's stool. The BSS scale is known to correlate well with transit time of food in the digestive system and identifies the presence of constipation and diarrhea, and precursors to colon cancer[2] .  \nFigure 1 . Bristol Stool Scale (BSS)  \nBy leveraging deep learning, and specifically convolutional neural networks (CNNs), we evaluated computer vision approaches that can augment physicians' efforts to classify stool images based on their BSS value. As part of this effort, we have obtained 886 images of stool in the wild. These images were collected from both internet searches and crowdsourced images from our website ([http://train.auggi.ai](http://train.auggi.ai)). The majority of the images were taken above a toilet with the camera aimed straight down  \nat the stool. Each image in our crea","cbCairl0jOQDBRP8","https://ap.wps.com/l/cbCairl0jOQDBRP8","pdf",778034,9,"English","# Abstract\n# 1. Introduction\n## Bristol Stool Scale and clinical motivation\n## Dataset collection and labeling\n## Previous segmentation work and consolidated BSS\n## Proposed approach and experiment design","[{\"question\":\"What problem does the paper address?\",\"answer\":\"The paper targets the lack of convenient, accurate measurement tools for chronic gut disorders by using stool images to extract standardized clinical data.\"},{\"question\":\"How is the Bristol Stool Scale (BSS) used in the system?\",\"answer\":\"Physicians assign a BSS score on a 1–7 scale based on visual assessment, and the model predicts BSS to quantify stool consistency over time.\"},{\"question\":\"What models and training strategies are evaluated?\",\"answer\":\"A ResNet18 baseline is trained to predict BSS, and deep metric learning is tested using triplet and contrastive loss by turning the CNN into a feature extractor and optimizing embeddings.\"}]","Stool Classification Using Deep Metric Learning | PDF",1791860213]